Symphony: Symmetry-Equivariant Point-Centered Spherical Harmonics for Molecule Generation

arXiv:2311.16199v2 Announce Type: replace
Abstract: We present Symphony, an $E(3)$-equivariant autoregressive generative model for 3D molecular geometries that iteratively builds a molecule from molecular fragments. Existing autoregressive models such as G-SchNet and G-SphereNet for molecules utilize rotationally invariant features to respect the 3D symmetries of molecules. In contrast, Symphony uses message-passing with higher-degree $E(3)$-equivariant features. This allows a novel representation of probability distributions via spherical harmonic signals to efficiently model the 3D geometry of molecules. We show that Symphony is …

abstract arxiv autoregressive models contrast cs.lg features generative molecules q-bio.bm symmetry type